Doug / smarc-fsl-linux-kernel

Download as
Browse Code ยป

Commit d5bdae7d59451b9d63303f7794ef32bb76ba6330

Authored by Glauber Costa
Committed by Linus Torvalds
1 parent 2ad306b17c
Exists in smarc-l5.0.0_1.0.0-ga and in 5 other branches imx_3.10.17_1.0.1_ga, imx_3.10.53_1.1.0_ga, imx_3.14.28_1.0.0_ga, smarc-imx_3.10.53_1.1.0_ga, smarc-imx_3.14.28_1.0.0_ga

memcg: add documentation about the kmem controller 

Signed-off-by: Glauber Costa <glommer@parallels.com>
Acked-by: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Frederic Weisbecker <fweisbec@redhat.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: JoonSoo Kim <js1304@gmail.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Rik van Riel <riel@redhat.com>
Cc: Suleiman Souhlal <suleiman@google.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Showing 1 changed file with 58 additions and 1 deletions Inline Diff

Documentation/cgroups/memory.txt
Diff comments   View file @ d5bdae7
1 Memory Resource Controller 1 Memory Resource Controller
2 2
3 NOTE: The Memory Resource Controller has generically been referred to as the 3 NOTE: The Memory Resource Controller has generically been referred to as the
4 memory controller in this document. Do not confuse memory controller 4 memory controller in this document. Do not confuse memory controller
5 used here with the memory controller that is used in hardware. 5 used here with the memory controller that is used in hardware.
6 6
7 (For editors) 7 (For editors)
8 In this document: 8 In this document:
9 When we mention a cgroup (cgroupfs's directory) with memory controller, 9 When we mention a cgroup (cgroupfs's directory) with memory controller,
10 we call it "memory cgroup". When you see git-log and source code, you'll 10 we call it "memory cgroup". When you see git-log and source code, you'll
11 see patch's title and function names tend to use "memcg". 11 see patch's title and function names tend to use "memcg".
12 In this document, we avoid using it. 12 In this document, we avoid using it.
13 13
14 Benefits and Purpose of the memory controller 14 Benefits and Purpose of the memory controller
15 15
16 The memory controller isolates the memory behaviour of a group of tasks 16 The memory controller isolates the memory behaviour of a group of tasks
17 from the rest of the system. The article on LWN [12] mentions some probable 17 from the rest of the system. The article on LWN [12] mentions some probable
18 uses of the memory controller. The memory controller can be used to 18 uses of the memory controller. The memory controller can be used to
19 19
20 a. Isolate an application or a group of applications 20 a. Isolate an application or a group of applications
21 Memory-hungry applications can be isolated and limited to a smaller 21 Memory-hungry applications can be isolated and limited to a smaller
22 amount of memory. 22 amount of memory.
23 b. Create a cgroup with a limited amount of memory; this can be used 23 b. Create a cgroup with a limited amount of memory; this can be used
24 as a good alternative to booting with mem=XXXX. 24 as a good alternative to booting with mem=XXXX.
25 c. Virtualization solutions can control the amount of memory they want 25 c. Virtualization solutions can control the amount of memory they want
26 to assign to a virtual machine instance. 26 to assign to a virtual machine instance.
27 d. A CD/DVD burner could control the amount of memory used by the 27 d. A CD/DVD burner could control the amount of memory used by the
28 rest of the system to ensure that burning does not fail due to lack 28 rest of the system to ensure that burning does not fail due to lack
29 of available memory. 29 of available memory.
30 e. There are several other use cases; find one or use the controller just 30 e. There are several other use cases; find one or use the controller just
31 for fun (to learn and hack on the VM subsystem). 31 for fun (to learn and hack on the VM subsystem).
32 32
33 Current Status: linux-2.6.34-mmotm(development version of 2010/April) 33 Current Status: linux-2.6.34-mmotm(development version of 2010/April)
34 34
35 Features: 35 Features:
36 - accounting anonymous pages, file caches, swap caches usage and limiting them. 36 - accounting anonymous pages, file caches, swap caches usage and limiting them.
37 - pages are linked to per-memcg LRU exclusively, and there is no global LRU. 37 - pages are linked to per-memcg LRU exclusively, and there is no global LRU.
38 - optionally, memory+swap usage can be accounted and limited. 38 - optionally, memory+swap usage can be accounted and limited.
39 - hierarchical accounting 39 - hierarchical accounting
40 - soft limit 40 - soft limit
41 - moving (recharging) account at moving a task is selectable. 41 - moving (recharging) account at moving a task is selectable.
42 - usage threshold notifier 42 - usage threshold notifier
43 - oom-killer disable knob and oom-notifier 43 - oom-killer disable knob and oom-notifier
44 - Root cgroup has no limit controls. 44 - Root cgroup has no limit controls.
45 45
46 Kernel memory support is a work in progress, and the current version provides 46 Kernel memory support is a work in progress, and the current version provides
47 basically functionality. (See Section 2.7) 47 basically functionality. (See Section 2.7)
48 48
49 Brief summary of control files. 49 Brief summary of control files.
50 50
51 tasks # attach a task(thread) and show list of threads 51 tasks # attach a task(thread) and show list of threads
52 cgroup.procs # show list of processes 52 cgroup.procs # show list of processes
53 cgroup.event_control # an interface for event_fd() 53 cgroup.event_control # an interface for event_fd()
54 memory.usage_in_bytes # show current res_counter usage for memory 54 memory.usage_in_bytes # show current res_counter usage for memory
55 (See 5.5 for details) 55 (See 5.5 for details)
56 memory.memsw.usage_in_bytes # show current res_counter usage for memory+Swap 56 memory.memsw.usage_in_bytes # show current res_counter usage for memory+Swap
57 (See 5.5 for details) 57 (See 5.5 for details)
58 memory.limit_in_bytes # set/show limit of memory usage 58 memory.limit_in_bytes # set/show limit of memory usage
59 memory.memsw.limit_in_bytes # set/show limit of memory+Swap usage 59 memory.memsw.limit_in_bytes # set/show limit of memory+Swap usage
60 memory.failcnt # show the number of memory usage hits limits 60 memory.failcnt # show the number of memory usage hits limits
61 memory.memsw.failcnt # show the number of memory+Swap hits limits 61 memory.memsw.failcnt # show the number of memory+Swap hits limits
62 memory.max_usage_in_bytes # show max memory usage recorded 62 memory.max_usage_in_bytes # show max memory usage recorded
63 memory.memsw.max_usage_in_bytes # show max memory+Swap usage recorded 63 memory.memsw.max_usage_in_bytes # show max memory+Swap usage recorded
64 memory.soft_limit_in_bytes # set/show soft limit of memory usage 64 memory.soft_limit_in_bytes # set/show soft limit of memory usage
65 memory.stat # show various statistics 65 memory.stat # show various statistics
66 memory.use_hierarchy # set/show hierarchical account enabled 66 memory.use_hierarchy # set/show hierarchical account enabled
67 memory.force_empty # trigger forced move charge to parent 67 memory.force_empty # trigger forced move charge to parent
68 memory.swappiness # set/show swappiness parameter of vmscan 68 memory.swappiness # set/show swappiness parameter of vmscan
69 (See sysctl's vm.swappiness) 69 (See sysctl's vm.swappiness)
70 memory.move_charge_at_immigrate # set/show controls of moving charges 70 memory.move_charge_at_immigrate # set/show controls of moving charges
71 memory.oom_control # set/show oom controls. 71 memory.oom_control # set/show oom controls.
72 memory.numa_stat # show the number of memory usage per numa node 72 memory.numa_stat # show the number of memory usage per numa node
73 73
74 memory.kmem.limit_in_bytes # set/show hard limit for kernel memory
75 memory.kmem.usage_in_bytes # show current kernel memory allocation
76 memory.kmem.failcnt # show the number of kernel memory usage hits limits
77 memory.kmem.max_usage_in_bytes # show max kernel memory usage recorded
78
74 memory.kmem.tcp.limit_in_bytes # set/show hard limit for tcp buf memory 79 memory.kmem.tcp.limit_in_bytes # set/show hard limit for tcp buf memory
75 memory.kmem.tcp.usage_in_bytes # show current tcp buf memory allocation 80 memory.kmem.tcp.usage_in_bytes # show current tcp buf memory allocation
76 memory.kmem.tcp.failcnt # show the number of tcp buf memory usage hits limits 81 memory.kmem.tcp.failcnt # show the number of tcp buf memory usage hits limits
77 memory.kmem.tcp.max_usage_in_bytes # show max tcp buf memory usage recorded 82 memory.kmem.tcp.max_usage_in_bytes # show max tcp buf memory usage recorded
78 83
79 1. History 84 1. History
80 85
81 The memory controller has a long history. A request for comments for the memory 86 The memory controller has a long history. A request for comments for the memory
82 controller was posted by Balbir Singh [1]. At the time the RFC was posted 87 controller was posted by Balbir Singh [1]. At the time the RFC was posted
83 there were several implementations for memory control. The goal of the 88 there were several implementations for memory control. The goal of the
84 RFC was to build consensus and agreement for the minimal features required 89 RFC was to build consensus and agreement for the minimal features required
85 for memory control. The first RSS controller was posted by Balbir Singh[2] 90 for memory control. The first RSS controller was posted by Balbir Singh[2]
86 in Feb 2007. Pavel Emelianov [3][4][5] has since posted three versions of the 91 in Feb 2007. Pavel Emelianov [3][4][5] has since posted three versions of the
87 RSS controller. At OLS, at the resource management BoF, everyone suggested 92 RSS controller. At OLS, at the resource management BoF, everyone suggested
88 that we handle both page cache and RSS together. Another request was raised 93 that we handle both page cache and RSS together. Another request was raised
89 to allow user space handling of OOM. The current memory controller is 94 to allow user space handling of OOM. The current memory controller is
90 at version 6; it combines both mapped (RSS) and unmapped Page 95 at version 6; it combines both mapped (RSS) and unmapped Page
91 Cache Control [11]. 96 Cache Control [11].
92 97
93 2. Memory Control 98 2. Memory Control
94 99
95 Memory is a unique resource in the sense that it is present in a limited 100 Memory is a unique resource in the sense that it is present in a limited
96 amount. If a task requires a lot of CPU processing, the task can spread 101 amount. If a task requires a lot of CPU processing, the task can spread
97 its processing over a period of hours, days, months or years, but with 102 its processing over a period of hours, days, months or years, but with
98 memory, the same physical memory needs to be reused to accomplish the task. 103 memory, the same physical memory needs to be reused to accomplish the task.
99 104
100 The memory controller implementation has been divided into phases. These 105 The memory controller implementation has been divided into phases. These
101 are: 106 are:
102 107
103 1. Memory controller 108 1. Memory controller
104 2. mlock(2) controller 109 2. mlock(2) controller
105 3. Kernel user memory accounting and slab control 110 3. Kernel user memory accounting and slab control
106 4. user mappings length controller 111 4. user mappings length controller
107 112
108 The memory controller is the first controller developed. 113 The memory controller is the first controller developed.
109 114
110 2.1. Design 115 2.1. Design
111 116
112 The core of the design is a counter called the res_counter. The res_counter 117 The core of the design is a counter called the res_counter. The res_counter
113 tracks the current memory usage and limit of the group of processes associated 118 tracks the current memory usage and limit of the group of processes associated
114 with the controller. Each cgroup has a memory controller specific data 119 with the controller. Each cgroup has a memory controller specific data
115 structure (mem_cgroup) associated with it. 120 structure (mem_cgroup) associated with it.
116 121
117 2.2. Accounting 122 2.2. Accounting
118 123
119 +--------------------+ 124 +--------------------+
120 | mem_cgroup | 125 | mem_cgroup |
121 | (res_counter) | 126 | (res_counter) |
122 +--------------------+ 127 +--------------------+
123 / ^ \ 128 / ^ \
124 / | \ 129 / | \
125 +---------------+ | +---------------+ 130 +---------------+ | +---------------+
126 | mm_struct | |.... | mm_struct | 131 | mm_struct | |.... | mm_struct |
127 | | | | | 132 | | | | |
128 +---------------+ | +---------------+ 133 +---------------+ | +---------------+
129 | 134 |
130 + --------------+ 135 + --------------+
131 | 136 |
132 +---------------+ +------+--------+ 137 +---------------+ +------+--------+
133 | page +----------> page_cgroup| 138 | page +----------> page_cgroup|
134 | | | | 139 | | | |
135 +---------------+ +---------------+ 140 +---------------+ +---------------+
136 141
137 (Figure 1: Hierarchy of Accounting) 142 (Figure 1: Hierarchy of Accounting)
138 143
139 144
140 Figure 1 shows the important aspects of the controller 145 Figure 1 shows the important aspects of the controller
141 146
142 1. Accounting happens per cgroup 147 1. Accounting happens per cgroup
143 2. Each mm_struct knows about which cgroup it belongs to 148 2. Each mm_struct knows about which cgroup it belongs to
144 3. Each page has a pointer to the page_cgroup, which in turn knows the 149 3. Each page has a pointer to the page_cgroup, which in turn knows the
145 cgroup it belongs to 150 cgroup it belongs to
146 151
147 The accounting is done as follows: mem_cgroup_charge_common() is invoked to 152 The accounting is done as follows: mem_cgroup_charge_common() is invoked to
148 set up the necessary data structures and check if the cgroup that is being 153 set up the necessary data structures and check if the cgroup that is being
149 charged is over its limit. If it is, then reclaim is invoked on the cgroup. 154 charged is over its limit. If it is, then reclaim is invoked on the cgroup.
150 More details can be found in the reclaim section of this document. 155 More details can be found in the reclaim section of this document.
151 If everything goes well, a page meta-data-structure called page_cgroup is 156 If everything goes well, a page meta-data-structure called page_cgroup is
152 updated. page_cgroup has its own LRU on cgroup. 157 updated. page_cgroup has its own LRU on cgroup.
153 (*) page_cgroup structure is allocated at boot/memory-hotplug time. 158 (*) page_cgroup structure is allocated at boot/memory-hotplug time.
154 159
155 2.2.1 Accounting details 160 2.2.1 Accounting details
156 161
157 All mapped anon pages (RSS) and cache pages (Page Cache) are accounted. 162 All mapped anon pages (RSS) and cache pages (Page Cache) are accounted.
158 Some pages which are never reclaimable and will not be on the LRU 163 Some pages which are never reclaimable and will not be on the LRU
159 are not accounted. We just account pages under usual VM management. 164 are not accounted. We just account pages under usual VM management.
160 165
161 RSS pages are accounted at page_fault unless they've already been accounted 166 RSS pages are accounted at page_fault unless they've already been accounted
162 for earlier. A file page will be accounted for as Page Cache when it's 167 for earlier. A file page will be accounted for as Page Cache when it's
163 inserted into inode (radix-tree). While it's mapped into the page tables of 168 inserted into inode (radix-tree). While it's mapped into the page tables of
164 processes, duplicate accounting is carefully avoided. 169 processes, duplicate accounting is carefully avoided.
165 170
166 An RSS page is unaccounted when it's fully unmapped. A PageCache page is 171 An RSS page is unaccounted when it's fully unmapped. A PageCache page is
167 unaccounted when it's removed from radix-tree. Even if RSS pages are fully 172 unaccounted when it's removed from radix-tree. Even if RSS pages are fully
168 unmapped (by kswapd), they may exist as SwapCache in the system until they 173 unmapped (by kswapd), they may exist as SwapCache in the system until they
169 are really freed. Such SwapCaches are also accounted. 174 are really freed. Such SwapCaches are also accounted.
170 A swapped-in page is not accounted until it's mapped. 175 A swapped-in page is not accounted until it's mapped.
171 176
172 Note: The kernel does swapin-readahead and reads multiple swaps at once. 177 Note: The kernel does swapin-readahead and reads multiple swaps at once.
173 This means swapped-in pages may contain pages for other tasks than a task 178 This means swapped-in pages may contain pages for other tasks than a task
174 causing page fault. So, we avoid accounting at swap-in I/O. 179 causing page fault. So, we avoid accounting at swap-in I/O.
175 180
176 At page migration, accounting information is kept. 181 At page migration, accounting information is kept.
177 182
178 Note: we just account pages-on-LRU because our purpose is to control amount 183 Note: we just account pages-on-LRU because our purpose is to control amount
179 of used pages; not-on-LRU pages tend to be out-of-control from VM view. 184 of used pages; not-on-LRU pages tend to be out-of-control from VM view.
180 185
181 2.3 Shared Page Accounting 186 2.3 Shared Page Accounting
182 187
183 Shared pages are accounted on the basis of the first touch approach. The 188 Shared pages are accounted on the basis of the first touch approach. The
184 cgroup that first touches a page is accounted for the page. The principle 189 cgroup that first touches a page is accounted for the page. The principle
185 behind this approach is that a cgroup that aggressively uses a shared 190 behind this approach is that a cgroup that aggressively uses a shared
186 page will eventually get charged for it (once it is uncharged from 191 page will eventually get charged for it (once it is uncharged from
187 the cgroup that brought it in -- this will happen on memory pressure). 192 the cgroup that brought it in -- this will happen on memory pressure).
188 193
189 But see section 8.2: when moving a task to another cgroup, its pages may 194 But see section 8.2: when moving a task to another cgroup, its pages may
190 be recharged to the new cgroup, if move_charge_at_immigrate has been chosen. 195 be recharged to the new cgroup, if move_charge_at_immigrate has been chosen.
191 196
192 Exception: If CONFIG_CGROUP_CGROUP_MEMCG_SWAP is not used. 197 Exception: If CONFIG_CGROUP_CGROUP_MEMCG_SWAP is not used.
193 When you do swapoff and make swapped-out pages of shmem(tmpfs) to 198 When you do swapoff and make swapped-out pages of shmem(tmpfs) to
194 be backed into memory in force, charges for pages are accounted against the 199 be backed into memory in force, charges for pages are accounted against the
195 caller of swapoff rather than the users of shmem. 200 caller of swapoff rather than the users of shmem.
196 201
197 2.4 Swap Extension (CONFIG_MEMCG_SWAP) 202 2.4 Swap Extension (CONFIG_MEMCG_SWAP)
198 203
199 Swap Extension allows you to record charge for swap. A swapped-in page is 204 Swap Extension allows you to record charge for swap. A swapped-in page is
200 charged back to original page allocator if possible. 205 charged back to original page allocator if possible.
201 206
202 When swap is accounted, following files are added. 207 When swap is accounted, following files are added.
203 - memory.memsw.usage_in_bytes. 208 - memory.memsw.usage_in_bytes.
204 - memory.memsw.limit_in_bytes. 209 - memory.memsw.limit_in_bytes.
205 210
206 memsw means memory+swap. Usage of memory+swap is limited by 211 memsw means memory+swap. Usage of memory+swap is limited by
207 memsw.limit_in_bytes. 212 memsw.limit_in_bytes.
208 213
209 Example: Assume a system with 4G of swap. A task which allocates 6G of memory 214 Example: Assume a system with 4G of swap. A task which allocates 6G of memory
210 (by mistake) under 2G memory limitation will use all swap. 215 (by mistake) under 2G memory limitation will use all swap.
211 In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap. 216 In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap.
212 By using the memsw limit, you can avoid system OOM which can be caused by swap 217 By using the memsw limit, you can avoid system OOM which can be caused by swap
213 shortage. 218 shortage.
214 219
215 * why 'memory+swap' rather than swap. 220 * why 'memory+swap' rather than swap.
216 The global LRU(kswapd) can swap out arbitrary pages. Swap-out means 221 The global LRU(kswapd) can swap out arbitrary pages. Swap-out means
217 to move account from memory to swap...there is no change in usage of 222 to move account from memory to swap...there is no change in usage of
218 memory+swap. In other words, when we want to limit the usage of swap without 223 memory+swap. In other words, when we want to limit the usage of swap without
219 affecting global LRU, memory+swap limit is better than just limiting swap from 224 affecting global LRU, memory+swap limit is better than just limiting swap from
220 an OS point of view. 225 an OS point of view.
221 226
222 * What happens when a cgroup hits memory.memsw.limit_in_bytes 227 * What happens when a cgroup hits memory.memsw.limit_in_bytes
223 When a cgroup hits memory.memsw.limit_in_bytes, it's useless to do swap-out 228 When a cgroup hits memory.memsw.limit_in_bytes, it's useless to do swap-out
224 in this cgroup. Then, swap-out will not be done by cgroup routine and file 229 in this cgroup. Then, swap-out will not be done by cgroup routine and file
225 caches are dropped. But as mentioned above, global LRU can do swapout memory 230 caches are dropped. But as mentioned above, global LRU can do swapout memory
226 from it for sanity of the system's memory management state. You can't forbid 231 from it for sanity of the system's memory management state. You can't forbid
227 it by cgroup. 232 it by cgroup.
228 233
229 2.5 Reclaim 234 2.5 Reclaim
230 235
231 Each cgroup maintains a per cgroup LRU which has the same structure as 236 Each cgroup maintains a per cgroup LRU which has the same structure as
232 global VM. When a cgroup goes over its limit, we first try 237 global VM. When a cgroup goes over its limit, we first try
233 to reclaim memory from the cgroup so as to make space for the new 238 to reclaim memory from the cgroup so as to make space for the new
234 pages that the cgroup has touched. If the reclaim is unsuccessful, 239 pages that the cgroup has touched. If the reclaim is unsuccessful,
235 an OOM routine is invoked to select and kill the bulkiest task in the 240 an OOM routine is invoked to select and kill the bulkiest task in the
236 cgroup. (See 10. OOM Control below.) 241 cgroup. (See 10. OOM Control below.)
237 242
238 The reclaim algorithm has not been modified for cgroups, except that 243 The reclaim algorithm has not been modified for cgroups, except that
239 pages that are selected for reclaiming come from the per-cgroup LRU 244 pages that are selected for reclaiming come from the per-cgroup LRU
240 list. 245 list.
241 246
242 NOTE: Reclaim does not work for the root cgroup, since we cannot set any 247 NOTE: Reclaim does not work for the root cgroup, since we cannot set any
243 limits on the root cgroup. 248 limits on the root cgroup.
244 249
245 Note2: When panic_on_oom is set to "2", the whole system will panic. 250 Note2: When panic_on_oom is set to "2", the whole system will panic.
246 251
247 When oom event notifier is registered, event will be delivered. 252 When oom event notifier is registered, event will be delivered.
248 (See oom_control section) 253 (See oom_control section)
249 254
250 2.6 Locking 255 2.6 Locking
251 256
252 lock_page_cgroup()/unlock_page_cgroup() should not be called under 257 lock_page_cgroup()/unlock_page_cgroup() should not be called under
253 mapping->tree_lock. 258 mapping->tree_lock.
254 259
255 Other lock order is following: 260 Other lock order is following:
256 PG_locked. 261 PG_locked.
257 mm->page_table_lock 262 mm->page_table_lock
258 zone->lru_lock 263 zone->lru_lock
259 lock_page_cgroup. 264 lock_page_cgroup.
260 In many cases, just lock_page_cgroup() is called. 265 In many cases, just lock_page_cgroup() is called.
261 per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by 266 per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by
262 zone->lru_lock, it has no lock of its own. 267 zone->lru_lock, it has no lock of its own.
263 268
264 2.7 Kernel Memory Extension (CONFIG_MEMCG_KMEM) 269 2.7 Kernel Memory Extension (CONFIG_MEMCG_KMEM)
265 270
266 With the Kernel memory extension, the Memory Controller is able to limit 271 With the Kernel memory extension, the Memory Controller is able to limit
267 the amount of kernel memory used by the system. Kernel memory is fundamentally 272 the amount of kernel memory used by the system. Kernel memory is fundamentally
268 different than user memory, since it can't be swapped out, which makes it 273 different than user memory, since it can't be swapped out, which makes it
269 possible to DoS the system by consuming too much of this precious resource. 274 possible to DoS the system by consuming too much of this precious resource.
270 275
276 Kernel memory won't be accounted at all until limit on a group is set. This
277 allows for existing setups to continue working without disruption. The limit
278 cannot be set if the cgroup have children, or if there are already tasks in the
279 cgroup. Attempting to set the limit under those conditions will return -EBUSY.
280 When use_hierarchy == 1 and a group is accounted, its children will
281 automatically be accounted regardless of their limit value.
282
283 After a group is first limited, it will be kept being accounted until it
284 is removed. The memory limitation itself, can of course be removed by writing
285 -1 to memory.kmem.limit_in_bytes. In this case, kmem will be accounted, but not
286 limited.
287
271 Kernel memory limits are not imposed for the root cgroup. Usage for the root 288 Kernel memory limits are not imposed for the root cgroup. Usage for the root
272 cgroup may or may not be accounted. 289 cgroup may or may not be accounted. The memory used is accumulated into
290 memory.kmem.usage_in_bytes, or in a separate counter when it makes sense.
291 (currently only for tcp).
292 The main "kmem" counter is fed into the main counter, so kmem charges will
293 also be visible from the user counter.
273 294
274 Currently no soft limit is implemented for kernel memory. It is future work 295 Currently no soft limit is implemented for kernel memory. It is future work
275 to trigger slab reclaim when those limits are reached. 296 to trigger slab reclaim when those limits are reached.
276 297
277 2.7.1 Current Kernel Memory resources accounted 298 2.7.1 Current Kernel Memory resources accounted
278 299
300 * stack pages: every process consumes some stack pages. By accounting into
301 kernel memory, we prevent new processes from being created when the kernel
302 memory usage is too high.
303
279 * sockets memory pressure: some sockets protocols have memory pressure 304 * sockets memory pressure: some sockets protocols have memory pressure
280 thresholds. The Memory Controller allows them to be controlled individually 305 thresholds. The Memory Controller allows them to be controlled individually
281 per cgroup, instead of globally. 306 per cgroup, instead of globally.
282 307
283 * tcp memory pressure: sockets memory pressure for the tcp protocol. 308 * tcp memory pressure: sockets memory pressure for the tcp protocol.
284 309
310 2.7.3 Common use cases
311
312 Because the "kmem" counter is fed to the main user counter, kernel memory can
313 never be limited completely independently of user memory. Say "U" is the user
314 limit, and "K" the kernel limit. There are three possible ways limits can be
315 set:
316
317 U != 0, K = unlimited:
318 This is the standard memcg limitation mechanism already present before kmem
319 accounting. Kernel memory is completely ignored.
320
321 U != 0, K < U:
322 Kernel memory is a subset of the user memory. This setup is useful in
323 deployments where the total amount of memory per-cgroup is overcommited.
324 Overcommiting kernel memory limits is definitely not recommended, since the
325 box can still run out of non-reclaimable memory.
326 In this case, the admin could set up K so that the sum of all groups is
327 never greater than the total memory, and freely set U at the cost of his
328 QoS.
329
330 U != 0, K >= U:
331 Since kmem charges will also be fed to the user counter and reclaim will be
332 triggered for the cgroup for both kinds of memory. This setup gives the
333 admin a unified view of memory, and it is also useful for people who just
334 want to track kernel memory usage.
335
285 3. User Interface 336 3. User Interface
286 337
287 0. Configuration 338 0. Configuration
288 339
289 a. Enable CONFIG_CGROUPS 340 a. Enable CONFIG_CGROUPS
290 b. Enable CONFIG_RESOURCE_COUNTERS 341 b. Enable CONFIG_RESOURCE_COUNTERS
291 c. Enable CONFIG_MEMCG 342 c. Enable CONFIG_MEMCG
292 d. Enable CONFIG_MEMCG_SWAP (to use swap extension) 343 d. Enable CONFIG_MEMCG_SWAP (to use swap extension)
344 d. Enable CONFIG_MEMCG_KMEM (to use kmem extension)
293 345
294 1. Prepare the cgroups (see cgroups.txt, Why are cgroups needed?) 346 1. Prepare the cgroups (see cgroups.txt, Why are cgroups needed?)
295 # mount -t tmpfs none /sys/fs/cgroup 347 # mount -t tmpfs none /sys/fs/cgroup
296 # mkdir /sys/fs/cgroup/memory 348 # mkdir /sys/fs/cgroup/memory
297 # mount -t cgroup none /sys/fs/cgroup/memory -o memory 349 # mount -t cgroup none /sys/fs/cgroup/memory -o memory
298 350
299 2. Make the new group and move bash into it 351 2. Make the new group and move bash into it
300 # mkdir /sys/fs/cgroup/memory/0 352 # mkdir /sys/fs/cgroup/memory/0
301 # echo $$ > /sys/fs/cgroup/memory/0/tasks 353 # echo $$ > /sys/fs/cgroup/memory/0/tasks
302 354
303 Since now we're in the 0 cgroup, we can alter the memory limit: 355 Since now we're in the 0 cgroup, we can alter the memory limit:
304 # echo 4M > /sys/fs/cgroup/memory/0/memory.limit_in_bytes 356 # echo 4M > /sys/fs/cgroup/memory/0/memory.limit_in_bytes
305 357
306 NOTE: We can use a suffix (k, K, m, M, g or G) to indicate values in kilo, 358 NOTE: We can use a suffix (k, K, m, M, g or G) to indicate values in kilo,
307 mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes, Gibibytes.) 359 mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes, Gibibytes.)
308 360
309 NOTE: We can write "-1" to reset the *.limit_in_bytes(unlimited). 361 NOTE: We can write "-1" to reset the *.limit_in_bytes(unlimited).
310 NOTE: We cannot set limits on the root cgroup any more. 362 NOTE: We cannot set limits on the root cgroup any more.
311 363
312 # cat /sys/fs/cgroup/memory/0/memory.limit_in_bytes 364 # cat /sys/fs/cgroup/memory/0/memory.limit_in_bytes
313 4194304 365 4194304
314 366
315 We can check the usage: 367 We can check the usage:
316 # cat /sys/fs/cgroup/memory/0/memory.usage_in_bytes 368 # cat /sys/fs/cgroup/memory/0/memory.usage_in_bytes
317 1216512 369 1216512
318 370
319 A successful write to this file does not guarantee a successful setting of 371 A successful write to this file does not guarantee a successful setting of
320 this limit to the value written into the file. This can be due to a 372 this limit to the value written into the file. This can be due to a
321 number of factors, such as rounding up to page boundaries or the total 373 number of factors, such as rounding up to page boundaries or the total
322 availability of memory on the system. The user is required to re-read 374 availability of memory on the system. The user is required to re-read
323 this file after a write to guarantee the value committed by the kernel. 375 this file after a write to guarantee the value committed by the kernel.
324 376
325 # echo 1 > memory.limit_in_bytes 377 # echo 1 > memory.limit_in_bytes
326 # cat memory.limit_in_bytes 378 # cat memory.limit_in_bytes
327 4096 379 4096
328 380
329 The memory.failcnt field gives the number of times that the cgroup limit was 381 The memory.failcnt field gives the number of times that the cgroup limit was
330 exceeded. 382 exceeded.
331 383
332 The memory.stat file gives accounting information. Now, the number of 384 The memory.stat file gives accounting information. Now, the number of
333 caches, RSS and Active pages/Inactive pages are shown. 385 caches, RSS and Active pages/Inactive pages are shown.
334 386
335 4. Testing 387 4. Testing
336 388
337 For testing features and implementation, see memcg_test.txt. 389 For testing features and implementation, see memcg_test.txt.
338 390
339 Performance test is also important. To see pure memory controller's overhead, 391 Performance test is also important. To see pure memory controller's overhead,
340 testing on tmpfs will give you good numbers of small overheads. 392 testing on tmpfs will give you good numbers of small overheads.
341 Example: do kernel make on tmpfs. 393 Example: do kernel make on tmpfs.
342 394
343 Page-fault scalability is also important. At measuring parallel 395 Page-fault scalability is also important. At measuring parallel
344 page fault test, multi-process test may be better than multi-thread 396 page fault test, multi-process test may be better than multi-thread
345 test because it has noise of shared objects/status. 397 test because it has noise of shared objects/status.
346 398
347 But the above two are testing extreme situations. 399 But the above two are testing extreme situations.
348 Trying usual test under memory controller is always helpful. 400 Trying usual test under memory controller is always helpful.
349 401
350 4.1 Troubleshooting 402 4.1 Troubleshooting
351 403
352 Sometimes a user might find that the application under a cgroup is 404 Sometimes a user might find that the application under a cgroup is
353 terminated by the OOM killer. There are several causes for this: 405 terminated by the OOM killer. There are several causes for this:
354 406
355 1. The cgroup limit is too low (just too low to do anything useful) 407 1. The cgroup limit is too low (just too low to do anything useful)
356 2. The user is using anonymous memory and swap is turned off or too low 408 2. The user is using anonymous memory and swap is turned off or too low
357 409
358 A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of 410 A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of
359 some of the pages cached in the cgroup (page cache pages). 411 some of the pages cached in the cgroup (page cache pages).
360 412
361 To know what happens, disabling OOM_Kill as per "10. OOM Control" (below) and 413 To know what happens, disabling OOM_Kill as per "10. OOM Control" (below) and
362 seeing what happens will be helpful. 414 seeing what happens will be helpful.
363 415
364 4.2 Task migration 416 4.2 Task migration
365 417
366 When a task migrates from one cgroup to another, its charge is not 418 When a task migrates from one cgroup to another, its charge is not
367 carried forward by default. The pages allocated from the original cgroup still 419 carried forward by default. The pages allocated from the original cgroup still
368 remain charged to it, the charge is dropped when the page is freed or 420 remain charged to it, the charge is dropped when the page is freed or
369 reclaimed. 421 reclaimed.
370 422
371 You can move charges of a task along with task migration. 423 You can move charges of a task along with task migration.
372 See 8. "Move charges at task migration" 424 See 8. "Move charges at task migration"
373 425
374 4.3 Removing a cgroup 426 4.3 Removing a cgroup
375 427
376 A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a 428 A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a
377 cgroup might have some charge associated with it, even though all 429 cgroup might have some charge associated with it, even though all
378 tasks have migrated away from it. (because we charge against pages, not 430 tasks have migrated away from it. (because we charge against pages, not
379 against tasks.) 431 against tasks.)
380 432
381 We move the stats to root (if use_hierarchy==0) or parent (if 433 We move the stats to root (if use_hierarchy==0) or parent (if
382 use_hierarchy==1), and no change on the charge except uncharging 434 use_hierarchy==1), and no change on the charge except uncharging
383 from the child. 435 from the child.
384 436
385 Charges recorded in swap information is not updated at removal of cgroup. 437 Charges recorded in swap information is not updated at removal of cgroup.
386 Recorded information is discarded and a cgroup which uses swap (swapcache) 438 Recorded information is discarded and a cgroup which uses swap (swapcache)
387 will be charged as a new owner of it. 439 will be charged as a new owner of it.
388 440
389 About use_hierarchy, see Section 6. 441 About use_hierarchy, see Section 6.
390 442
391 5. Misc. interfaces. 443 5. Misc. interfaces.
392 444
393 5.1 force_empty 445 5.1 force_empty
394 memory.force_empty interface is provided to make cgroup's memory usage empty. 446 memory.force_empty interface is provided to make cgroup's memory usage empty.
395 You can use this interface only when the cgroup has no tasks. 447 You can use this interface only when the cgroup has no tasks.
396 When writing anything to this 448 When writing anything to this
397 449
398 # echo 0 > memory.force_empty 450 # echo 0 > memory.force_empty
399 451
400 Almost all pages tracked by this memory cgroup will be unmapped and freed. 452 Almost all pages tracked by this memory cgroup will be unmapped and freed.
401 Some pages cannot be freed because they are locked or in-use. Such pages are 453 Some pages cannot be freed because they are locked or in-use. Such pages are
402 moved to parent (if use_hierarchy==1) or root (if use_hierarchy==0) and this 454 moved to parent (if use_hierarchy==1) or root (if use_hierarchy==0) and this
403 cgroup will be empty. 455 cgroup will be empty.
404 456
405 The typical use case for this interface is before calling rmdir(). 457 The typical use case for this interface is before calling rmdir().
406 Because rmdir() moves all pages to parent, some out-of-use page caches can be 458 Because rmdir() moves all pages to parent, some out-of-use page caches can be
407 moved to the parent. If you want to avoid that, force_empty will be useful. 459 moved to the parent. If you want to avoid that, force_empty will be useful.
460
461 Also, note that when memory.kmem.limit_in_bytes is set the charges due to
462 kernel pages will still be seen. This is not considered a failure and the
463 write will still return success. In this case, it is expected that
464 memory.kmem.usage_in_bytes == memory.usage_in_bytes.
408 465
409 About use_hierarchy, see Section 6. 466 About use_hierarchy, see Section 6.
410 467
411 5.2 stat file 468 5.2 stat file
412 469
413 memory.stat file includes following statistics 470 memory.stat file includes following statistics
414 471
415 # per-memory cgroup local status 472 # per-memory cgroup local status
416 cache - # of bytes of page cache memory. 473 cache - # of bytes of page cache memory.
417 rss - # of bytes of anonymous and swap cache memory. 474 rss - # of bytes of anonymous and swap cache memory.
418 mapped_file - # of bytes of mapped file (includes tmpfs/shmem) 475 mapped_file - # of bytes of mapped file (includes tmpfs/shmem)
419 pgpgin - # of charging events to the memory cgroup. The charging 476 pgpgin - # of charging events to the memory cgroup. The charging
420 event happens each time a page is accounted as either mapped 477 event happens each time a page is accounted as either mapped
421 anon page(RSS) or cache page(Page Cache) to the cgroup. 478 anon page(RSS) or cache page(Page Cache) to the cgroup.
422 pgpgout - # of uncharging events to the memory cgroup. The uncharging 479 pgpgout - # of uncharging events to the memory cgroup. The uncharging
423 event happens each time a page is unaccounted from the cgroup. 480 event happens each time a page is unaccounted from the cgroup.
424 swap - # of bytes of swap usage 481 swap - # of bytes of swap usage
425 inactive_anon - # of bytes of anonymous memory and swap cache memory on 482 inactive_anon - # of bytes of anonymous memory and swap cache memory on
426 LRU list. 483 LRU list.
427 active_anon - # of bytes of anonymous and swap cache memory on active 484 active_anon - # of bytes of anonymous and swap cache memory on active
428 inactive LRU list. 485 inactive LRU list.
429 inactive_file - # of bytes of file-backed memory on inactive LRU list. 486 inactive_file - # of bytes of file-backed memory on inactive LRU list.
430 active_file - # of bytes of file-backed memory on active LRU list. 487 active_file - # of bytes of file-backed memory on active LRU list.
431 unevictable - # of bytes of memory that cannot be reclaimed (mlocked etc). 488 unevictable - # of bytes of memory that cannot be reclaimed (mlocked etc).
432 489
433 # status considering hierarchy (see memory.use_hierarchy settings) 490 # status considering hierarchy (see memory.use_hierarchy settings)
434 491
435 hierarchical_memory_limit - # of bytes of memory limit with regard to hierarchy 492 hierarchical_memory_limit - # of bytes of memory limit with regard to hierarchy
436 under which the memory cgroup is 493 under which the memory cgroup is
437 hierarchical_memsw_limit - # of bytes of memory+swap limit with regard to 494 hierarchical_memsw_limit - # of bytes of memory+swap limit with regard to
438 hierarchy under which memory cgroup is. 495 hierarchy under which memory cgroup is.
439 496
440 total_<counter> - # hierarchical version of <counter>, which in 497 total_<counter> - # hierarchical version of <counter>, which in
441 addition to the cgroup's own value includes the 498 addition to the cgroup's own value includes the
442 sum of all hierarchical children's values of 499 sum of all hierarchical children's values of
443 <counter>, i.e. total_cache 500 <counter>, i.e. total_cache
444 501
445 # The following additional stats are dependent on CONFIG_DEBUG_VM. 502 # The following additional stats are dependent on CONFIG_DEBUG_VM.
446 503
447 recent_rotated_anon - VM internal parameter. (see mm/vmscan.c) 504 recent_rotated_anon - VM internal parameter. (see mm/vmscan.c)
448 recent_rotated_file - VM internal parameter. (see mm/vmscan.c) 505 recent_rotated_file - VM internal parameter. (see mm/vmscan.c)
449 recent_scanned_anon - VM internal parameter. (see mm/vmscan.c) 506 recent_scanned_anon - VM internal parameter. (see mm/vmscan.c)
450 recent_scanned_file - VM internal parameter. (see mm/vmscan.c) 507 recent_scanned_file - VM internal parameter. (see mm/vmscan.c)
451 508
452 Memo: 509 Memo:
453 recent_rotated means recent frequency of LRU rotation. 510 recent_rotated means recent frequency of LRU rotation.
454 recent_scanned means recent # of scans to LRU. 511 recent_scanned means recent # of scans to LRU.
455 showing for better debug please see the code for meanings. 512 showing for better debug please see the code for meanings.
456 513
457 Note: 514 Note:
458 Only anonymous and swap cache memory is listed as part of 'rss' stat. 515 Only anonymous and swap cache memory is listed as part of 'rss' stat.
459 This should not be confused with the true 'resident set size' or the 516 This should not be confused with the true 'resident set size' or the
460 amount of physical memory used by the cgroup. 517 amount of physical memory used by the cgroup.
461 'rss + file_mapped" will give you resident set size of cgroup. 518 'rss + file_mapped" will give you resident set size of cgroup.
462 (Note: file and shmem may be shared among other cgroups. In that case, 519 (Note: file and shmem may be shared among other cgroups. In that case,
463 file_mapped is accounted only when the memory cgroup is owner of page 520 file_mapped is accounted only when the memory cgroup is owner of page
464 cache.) 521 cache.)
465 522
466 5.3 swappiness 523 5.3 swappiness
467 524
468 Similar to /proc/sys/vm/swappiness, but affecting a hierarchy of groups only. 525 Similar to /proc/sys/vm/swappiness, but affecting a hierarchy of groups only.
469 Please note that unlike the global swappiness, memcg knob set to 0 526 Please note that unlike the global swappiness, memcg knob set to 0
470 really prevents from any swapping even if there is a swap storage 527 really prevents from any swapping even if there is a swap storage
471 available. This might lead to memcg OOM killer if there are no file 528 available. This might lead to memcg OOM killer if there are no file
472 pages to reclaim. 529 pages to reclaim.
473 530
474 Following cgroups' swappiness can't be changed. 531 Following cgroups' swappiness can't be changed.
475 - root cgroup (uses /proc/sys/vm/swappiness). 532 - root cgroup (uses /proc/sys/vm/swappiness).
476 - a cgroup which uses hierarchy and it has other cgroup(s) below it. 533 - a cgroup which uses hierarchy and it has other cgroup(s) below it.
477 - a cgroup which uses hierarchy and not the root of hierarchy. 534 - a cgroup which uses hierarchy and not the root of hierarchy.
478 535
479 5.4 failcnt 536 5.4 failcnt
480 537
481 A memory cgroup provides memory.failcnt and memory.memsw.failcnt files. 538 A memory cgroup provides memory.failcnt and memory.memsw.failcnt files.
482 This failcnt(== failure count) shows the number of times that a usage counter 539 This failcnt(== failure count) shows the number of times that a usage counter
483 hit its limit. When a memory cgroup hits a limit, failcnt increases and 540 hit its limit. When a memory cgroup hits a limit, failcnt increases and
484 memory under it will be reclaimed. 541 memory under it will be reclaimed.
485 542
486 You can reset failcnt by writing 0 to failcnt file. 543 You can reset failcnt by writing 0 to failcnt file.
487 # echo 0 > .../memory.failcnt 544 # echo 0 > .../memory.failcnt
488 545
489 5.5 usage_in_bytes 546 5.5 usage_in_bytes
490 547
491 For efficiency, as other kernel components, memory cgroup uses some optimization 548 For efficiency, as other kernel components, memory cgroup uses some optimization
492 to avoid unnecessary cacheline false sharing. usage_in_bytes is affected by the 549 to avoid unnecessary cacheline false sharing. usage_in_bytes is affected by the
493 method and doesn't show 'exact' value of memory (and swap) usage, it's a fuzz 550 method and doesn't show 'exact' value of memory (and swap) usage, it's a fuzz
494 value for efficient access. (Of course, when necessary, it's synchronized.) 551 value for efficient access. (Of course, when necessary, it's synchronized.)
495 If you want to know more exact memory usage, you should use RSS+CACHE(+SWAP) 552 If you want to know more exact memory usage, you should use RSS+CACHE(+SWAP)
496 value in memory.stat(see 5.2). 553 value in memory.stat(see 5.2).
497 554
498 5.6 numa_stat 555 5.6 numa_stat
499 556
500 This is similar to numa_maps but operates on a per-memcg basis. This is 557 This is similar to numa_maps but operates on a per-memcg basis. This is
501 useful for providing visibility into the numa locality information within 558 useful for providing visibility into the numa locality information within
502 an memcg since the pages are allowed to be allocated from any physical 559 an memcg since the pages are allowed to be allocated from any physical
503 node. One of the use cases is evaluating application performance by 560 node. One of the use cases is evaluating application performance by
504 combining this information with the application's CPU allocation. 561 combining this information with the application's CPU allocation.
505 562
506 We export "total", "file", "anon" and "unevictable" pages per-node for 563 We export "total", "file", "anon" and "unevictable" pages per-node for
507 each memcg. The ouput format of memory.numa_stat is: 564 each memcg. The ouput format of memory.numa_stat is:
508 565
509 total=<total pages> N0=<node 0 pages> N1=<node 1 pages> ... 566 total=<total pages> N0=<node 0 pages> N1=<node 1 pages> ...
510 file=<total file pages> N0=<node 0 pages> N1=<node 1 pages> ... 567 file=<total file pages> N0=<node 0 pages> N1=<node 1 pages> ...
511 anon=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ... 568 anon=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
512 unevictable=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ... 569 unevictable=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
513 570
514 And we have total = file + anon + unevictable. 571 And we have total = file + anon + unevictable.
515 572
516 6. Hierarchy support 573 6. Hierarchy support
517 574
518 The memory controller supports a deep hierarchy and hierarchical accounting. 575 The memory controller supports a deep hierarchy and hierarchical accounting.
519 The hierarchy is created by creating the appropriate cgroups in the 576 The hierarchy is created by creating the appropriate cgroups in the
520 cgroup filesystem. Consider for example, the following cgroup filesystem 577 cgroup filesystem. Consider for example, the following cgroup filesystem
521 hierarchy 578 hierarchy
522 579
523 root 580 root
524 / | \ 581 / | \
525 / | \ 582 / | \
526 a b c 583 a b c
527 | \ 584 | \
528 | \ 585 | \
529 d e 586 d e
530 587
531 In the diagram above, with hierarchical accounting enabled, all memory 588 In the diagram above, with hierarchical accounting enabled, all memory
532 usage of e, is accounted to its ancestors up until the root (i.e, c and root), 589 usage of e, is accounted to its ancestors up until the root (i.e, c and root),
533 that has memory.use_hierarchy enabled. If one of the ancestors goes over its 590 that has memory.use_hierarchy enabled. If one of the ancestors goes over its
534 limit, the reclaim algorithm reclaims from the tasks in the ancestor and the 591 limit, the reclaim algorithm reclaims from the tasks in the ancestor and the
535 children of the ancestor. 592 children of the ancestor.
536 593
537 6.1 Enabling hierarchical accounting and reclaim 594 6.1 Enabling hierarchical accounting and reclaim
538 595
539 A memory cgroup by default disables the hierarchy feature. Support 596 A memory cgroup by default disables the hierarchy feature. Support
540 can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup 597 can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup
541 598
542 # echo 1 > memory.use_hierarchy 599 # echo 1 > memory.use_hierarchy
543 600
544 The feature can be disabled by 601 The feature can be disabled by
545 602
546 # echo 0 > memory.use_hierarchy 603 # echo 0 > memory.use_hierarchy
547 604
548 NOTE1: Enabling/disabling will fail if either the cgroup already has other 605 NOTE1: Enabling/disabling will fail if either the cgroup already has other
549 cgroups created below it, or if the parent cgroup has use_hierarchy 606 cgroups created below it, or if the parent cgroup has use_hierarchy
550 enabled. 607 enabled.
551 608
552 NOTE2: When panic_on_oom is set to "2", the whole system will panic in 609 NOTE2: When panic_on_oom is set to "2", the whole system will panic in
553 case of an OOM event in any cgroup. 610 case of an OOM event in any cgroup.
554 611
555 7. Soft limits 612 7. Soft limits
556 613
557 Soft limits allow for greater sharing of memory. The idea behind soft limits 614 Soft limits allow for greater sharing of memory. The idea behind soft limits
558 is to allow control groups to use as much of the memory as needed, provided 615 is to allow control groups to use as much of the memory as needed, provided
559 616
560 a. There is no memory contention 617 a. There is no memory contention
561 b. They do not exceed their hard limit 618 b. They do not exceed their hard limit
562 619
563 When the system detects memory contention or low memory, control groups 620 When the system detects memory contention or low memory, control groups
564 are pushed back to their soft limits. If the soft limit of each control 621 are pushed back to their soft limits. If the soft limit of each control
565 group is very high, they are pushed back as much as possible to make 622 group is very high, they are pushed back as much as possible to make
566 sure that one control group does not starve the others of memory. 623 sure that one control group does not starve the others of memory.
567 624
568 Please note that soft limits is a best-effort feature; it comes with 625 Please note that soft limits is a best-effort feature; it comes with
569 no guarantees, but it does its best to make sure that when memory is 626 no guarantees, but it does its best to make sure that when memory is
570 heavily contended for, memory is allocated based on the soft limit 627 heavily contended for, memory is allocated based on the soft limit
571 hints/setup. Currently soft limit based reclaim is set up such that 628 hints/setup. Currently soft limit based reclaim is set up such that
572 it gets invoked from balance_pgdat (kswapd). 629 it gets invoked from balance_pgdat (kswapd).
573 630
574 7.1 Interface 631 7.1 Interface
575 632
576 Soft limits can be setup by using the following commands (in this example we 633 Soft limits can be setup by using the following commands (in this example we
577 assume a soft limit of 256 MiB) 634 assume a soft limit of 256 MiB)
578 635
579 # echo 256M > memory.soft_limit_in_bytes 636 # echo 256M > memory.soft_limit_in_bytes
580 637
581 If we want to change this to 1G, we can at any time use 638 If we want to change this to 1G, we can at any time use
582 639
583 # echo 1G > memory.soft_limit_in_bytes 640 # echo 1G > memory.soft_limit_in_bytes
584 641
585 NOTE1: Soft limits take effect over a long period of time, since they involve 642 NOTE1: Soft limits take effect over a long period of time, since they involve
586 reclaiming memory for balancing between memory cgroups 643 reclaiming memory for balancing between memory cgroups
587 NOTE2: It is recommended to set the soft limit always below the hard limit, 644 NOTE2: It is recommended to set the soft limit always below the hard limit,
588 otherwise the hard limit will take precedence. 645 otherwise the hard limit will take precedence.
589 646
590 8. Move charges at task migration 647 8. Move charges at task migration
591 648
592 Users can move charges associated with a task along with task migration, that 649 Users can move charges associated with a task along with task migration, that
593 is, uncharge task's pages from the old cgroup and charge them to the new cgroup. 650 is, uncharge task's pages from the old cgroup and charge them to the new cgroup.
594 This feature is not supported in !CONFIG_MMU environments because of lack of 651 This feature is not supported in !CONFIG_MMU environments because of lack of
595 page tables. 652 page tables.
596 653
597 8.1 Interface 654 8.1 Interface
598 655
599 This feature is disabled by default. It can be enabledi (and disabled again) by 656 This feature is disabled by default. It can be enabledi (and disabled again) by
600 writing to memory.move_charge_at_immigrate of the destination cgroup. 657 writing to memory.move_charge_at_immigrate of the destination cgroup.
601 658
602 If you want to enable it: 659 If you want to enable it:
603 660
604 # echo (some positive value) > memory.move_charge_at_immigrate 661 # echo (some positive value) > memory.move_charge_at_immigrate
605 662
606 Note: Each bits of move_charge_at_immigrate has its own meaning about what type 663 Note: Each bits of move_charge_at_immigrate has its own meaning about what type
607 of charges should be moved. See 8.2 for details. 664 of charges should be moved. See 8.2 for details.
608 Note: Charges are moved only when you move mm->owner, in other words, 665 Note: Charges are moved only when you move mm->owner, in other words,
609 a leader of a thread group. 666 a leader of a thread group.
610 Note: If we cannot find enough space for the task in the destination cgroup, we 667 Note: If we cannot find enough space for the task in the destination cgroup, we
611 try to make space by reclaiming memory. Task migration may fail if we 668 try to make space by reclaiming memory. Task migration may fail if we
612 cannot make enough space. 669 cannot make enough space.
613 Note: It can take several seconds if you move charges much. 670 Note: It can take several seconds if you move charges much.
614 671
615 And if you want disable it again: 672 And if you want disable it again:
616 673
617 # echo 0 > memory.move_charge_at_immigrate 674 # echo 0 > memory.move_charge_at_immigrate
618 675
619 8.2 Type of charges which can be moved 676 8.2 Type of charges which can be moved
620 677
621 Each bit in move_charge_at_immigrate has its own meaning about what type of 678 Each bit in move_charge_at_immigrate has its own meaning about what type of
622 charges should be moved. But in any case, it must be noted that an account of 679 charges should be moved. But in any case, it must be noted that an account of
623 a page or a swap can be moved only when it is charged to the task's current 680 a page or a swap can be moved only when it is charged to the task's current
624 (old) memory cgroup. 681 (old) memory cgroup.
625 682
626 bit | what type of charges would be moved ? 683 bit | what type of charges would be moved ?
627 -----+------------------------------------------------------------------------ 684 -----+------------------------------------------------------------------------
628 0 | A charge of an anonymous page (or swap of it) used by the target task. 685 0 | A charge of an anonymous page (or swap of it) used by the target task.
629 | You must enable Swap Extension (see 2.4) to enable move of swap charges. 686 | You must enable Swap Extension (see 2.4) to enable move of swap charges.
630 -----+------------------------------------------------------------------------ 687 -----+------------------------------------------------------------------------
631 1 | A charge of file pages (normal file, tmpfs file (e.g. ipc shared memory) 688 1 | A charge of file pages (normal file, tmpfs file (e.g. ipc shared memory)
632 | and swaps of tmpfs file) mmapped by the target task. Unlike the case of 689 | and swaps of tmpfs file) mmapped by the target task. Unlike the case of
633 | anonymous pages, file pages (and swaps) in the range mmapped by the task 690 | anonymous pages, file pages (and swaps) in the range mmapped by the task
634 | will be moved even if the task hasn't done page fault, i.e. they might 691 | will be moved even if the task hasn't done page fault, i.e. they might
635 | not be the task's "RSS", but other task's "RSS" that maps the same file. 692 | not be the task's "RSS", but other task's "RSS" that maps the same file.
636 | And mapcount of the page is ignored (the page can be moved even if 693 | And mapcount of the page is ignored (the page can be moved even if
637 | page_mapcount(page) > 1). You must enable Swap Extension (see 2.4) to 694 | page_mapcount(page) > 1). You must enable Swap Extension (see 2.4) to
638 | enable move of swap charges. 695 | enable move of swap charges.
639 696
640 8.3 TODO 697 8.3 TODO
641 698
642 - All of moving charge operations are done under cgroup_mutex. It's not good 699 - All of moving charge operations are done under cgroup_mutex. It's not good
643 behavior to hold the mutex too long, so we may need some trick. 700 behavior to hold the mutex too long, so we may need some trick.
644 701
645 9. Memory thresholds 702 9. Memory thresholds
646 703
647 Memory cgroup implements memory thresholds using the cgroups notification 704 Memory cgroup implements memory thresholds using the cgroups notification
648 API (see cgroups.txt). It allows to register multiple memory and memsw 705 API (see cgroups.txt). It allows to register multiple memory and memsw
649 thresholds and gets notifications when it crosses. 706 thresholds and gets notifications when it crosses.
650 707
651 To register a threshold, an application must: 708 To register a threshold, an application must:
652 - create an eventfd using eventfd(2); 709 - create an eventfd using eventfd(2);
653 - open memory.usage_in_bytes or memory.memsw.usage_in_bytes; 710 - open memory.usage_in_bytes or memory.memsw.usage_in_bytes;
654 - write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to 711 - write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to
655 cgroup.event_control. 712 cgroup.event_control.
656 713
657 Application will be notified through eventfd when memory usage crosses 714 Application will be notified through eventfd when memory usage crosses
658 threshold in any direction. 715 threshold in any direction.
659 716
660 It's applicable for root and non-root cgroup. 717 It's applicable for root and non-root cgroup.
661 718
662 10. OOM Control 719 10. OOM Control
663 720
664 memory.oom_control file is for OOM notification and other controls. 721 memory.oom_control file is for OOM notification and other controls.
665 722
666 Memory cgroup implements OOM notifier using the cgroup notification 723 Memory cgroup implements OOM notifier using the cgroup notification
667 API (See cgroups.txt). It allows to register multiple OOM notification 724 API (See cgroups.txt). It allows to register multiple OOM notification
668 delivery and gets notification when OOM happens. 725 delivery and gets notification when OOM happens.
669 726
670 To register a notifier, an application must: 727 To register a notifier, an application must:
671 - create an eventfd using eventfd(2) 728 - create an eventfd using eventfd(2)
672 - open memory.oom_control file 729 - open memory.oom_control file
673 - write string like "<event_fd> <fd of memory.oom_control>" to 730 - write string like "<event_fd> <fd of memory.oom_control>" to
674 cgroup.event_control 731 cgroup.event_control
675 732
676 The application will be notified through eventfd when OOM happens. 733 The application will be notified through eventfd when OOM happens.
677 OOM notification doesn't work for the root cgroup. 734 OOM notification doesn't work for the root cgroup.
678 735
679 You can disable the OOM-killer by writing "1" to memory.oom_control file, as: 736 You can disable the OOM-killer by writing "1" to memory.oom_control file, as:
680 737
681 #echo 1 > memory.oom_control 738 #echo 1 > memory.oom_control
682 739
683 This operation is only allowed to the top cgroup of a sub-hierarchy. 740 This operation is only allowed to the top cgroup of a sub-hierarchy.
684 If OOM-killer is disabled, tasks under cgroup will hang/sleep 741 If OOM-killer is disabled, tasks under cgroup will hang/sleep
685 in memory cgroup's OOM-waitqueue when they request accountable memory. 742 in memory cgroup's OOM-waitqueue when they request accountable memory.
686 743
687 For running them, you have to relax the memory cgroup's OOM status by 744 For running them, you have to relax the memory cgroup's OOM status by
688 * enlarge limit or reduce usage. 745 * enlarge limit or reduce usage.
689 To reduce usage, 746 To reduce usage,
690 * kill some tasks. 747 * kill some tasks.
691 * move some tasks to other group with account migration. 748 * move some tasks to other group with account migration.
692 * remove some files (on tmpfs?) 749 * remove some files (on tmpfs?)
693 750
694 Then, stopped tasks will work again. 751 Then, stopped tasks will work again.
695 752
696 At reading, current status of OOM is shown. 753 At reading, current status of OOM is shown.
697 oom_kill_disable 0 or 1 (if 1, oom-killer is disabled) 754 oom_kill_disable 0 or 1 (if 1, oom-killer is disabled)
698 under_oom 0 or 1 (if 1, the memory cgroup is under OOM, tasks may 755 under_oom 0 or 1 (if 1, the memory cgroup is under OOM, tasks may
699 be stopped.) 756 be stopped.)
700 757
701 11. TODO 758 11. TODO
702 759
703 1. Add support for accounting huge pages (as a separate controller) 760 1. Add support for accounting huge pages (as a separate controller)
704 2. Make per-cgroup scanner reclaim not-shared pages first 761 2. Make per-cgroup scanner reclaim not-shared pages first
705 3. Teach controller to account for shared-pages 762 3. Teach controller to account for shared-pages
706 4. Start reclamation in the background when the limit is 763 4. Start reclamation in the background when the limit is
707 not yet hit but the usage is getting closer 764 not yet hit but the usage is getting closer
708 765
709 Summary 766 Summary
710 767
711 Overall, the memory controller has been a stable controller and has been 768 Overall, the memory controller has been a stable controller and has been
712 commented and discussed quite extensively in the community. 769 commented and discussed quite extensively in the community.
713 770
714 References 771 References
715 772
716 1. Singh, Balbir. RFC: Memory Controller, http://lwn.net/Articles/206697/ 773 1. Singh, Balbir. RFC: Memory Controller, http://lwn.net/Articles/206697/
717 2. Singh, Balbir. Memory Controller (RSS Control), 774 2. Singh, Balbir. Memory Controller (RSS Control),
718 http://lwn.net/Articles/222762/ 775 http://lwn.net/Articles/222762/
719 3. Emelianov, Pavel. Resource controllers based on process cgroups 776 3. Emelianov, Pavel. Resource controllers based on process cgroups
720 http://lkml.org/lkml/2007/3/6/198 777 http://lkml.org/lkml/2007/3/6/198
721 4. Emelianov, Pavel. RSS controller based on process cgroups (v2) 778 4. Emelianov, Pavel. RSS controller based on process cgroups (v2)
722 http://lkml.org/lkml/2007/4/9/78 779 http://lkml.org/lkml/2007/4/9/78
723 5. Emelianov, Pavel. RSS controller based on process cgroups (v3) 780 5. Emelianov, Pavel. RSS controller based on process cgroups (v3)
724 http://lkml.org/lkml/2007/5/30/244 781 http://lkml.org/lkml/2007/5/30/244
725 6. Menage, Paul. Control Groups v10, http://lwn.net/Articles/236032/ 782 6. Menage, Paul. Control Groups v10, http://lwn.net/Articles/236032/
726 7. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control 783 7. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control
727 subsystem (v3), http://lwn.net/Articles/235534/ 784 subsystem (v3), http://lwn.net/Articles/235534/
728 8. Singh, Balbir. RSS controller v2 test results (lmbench), 785 8. Singh, Balbir. RSS controller v2 test results (lmbench),
729 http://lkml.org/lkml/2007/5/17/232 786 http://lkml.org/lkml/2007/5/17/232
730 9. Singh, Balbir. RSS controller v2 AIM9 results 787 9. Singh, Balbir. RSS controller v2 AIM9 results
731 http://lkml.org/lkml/2007/5/18/1 788 http://lkml.org/lkml/2007/5/18/1
732 10. Singh, Balbir. Memory controller v6 test results, 789 10. Singh, Balbir. Memory controller v6 test results,
733 http://lkml.org/lkml/2007/8/19/36 790 http://lkml.org/lkml/2007/8/19/36
734 11. Singh, Balbir. Memory controller introduction (v6), 791 11. Singh, Balbir. Memory controller introduction (v6),
735 http://lkml.org/lkml/2007/8/17/69 792 http://lkml.org/lkml/2007/8/17/69
736 12. Corbet, Jonathan, Controlling memory use in cgroups, 793 12. Corbet, Jonathan, Controlling memory use in cgroups,
737 http://lwn.net/Articles/243795/ 794 http://lwn.net/Articles/243795/
738 795